Joint modeling of high-dimensional longitudinal data and survival using supervised low-rank tensor decomposition

Area of Science:

• Biostatistics
• Computational Biology
• Biomedical Informatics

Background:

• High-dimensional longitudinal data from omics platforms are common in biomedical research.
• Jointly modeling these data with survival outcomes presents challenges in temporal dynamics, feature dependencies, and computation.
• Existing methods struggle with the complexity and scale of modern biomedical datasets.

Purpose of the Study:

• To develop a novel joint modeling framework for high-dimensional longitudinal data and survival outcomes.
• To capture latent structure in longitudinal data and model time-to-event outcomes coherently.
• To enable accurate individualized predictions of disease progression and risk.

Main Methods:

• Utilized supervised low-rank functional tensor decomposition for multivariate longitudinal data.
• Employed proportional hazards modeling for time-to-event outcomes.
• Implemented a likelihood-based Monte Carlo Expectation-Maximization algorithm for estimation and prediction.

Main Results:

• The proposed framework effectively captures latent structure and dependencies in high-dimensional longitudinal data.
• Demonstrated substantial improvements in estimation accuracy and predictive performance over standard methods in simulations.
• Achieved over 99% variation explained in Alzheimer's Disease Neuroimaging Initiative lipidomics data, identifying dementia onset predictors.

Conclusions:

• The novel framework offers a scalable and interpretable strategy for integrating high-dimensional biomarkers.
• Provides dynamic predictions of longitudinal trajectories and survival probabilities for personalized risk assessment.
• Enhances joint modeling capabilities for complex biomedical data, advancing disease progression and risk stratification research.